Persistent memory for AI coding agents

Your AI remembers everything — architecture, decisions, patterns, progress — across every session, automatically. No prompts. No tool calls. Zero tokens spent on memory management.

$curl -fsSL https://codexfi.com/install | bash
codexfi memory flow — conversations are extracted into typed memories, stored in a vector database, and retrieved for future sessions[MEMORY] injectedarchitecturetech-contextprogresslearned-patternpreferenceOpenCode AgentImplement Auth flowusing NextAuth.jsI'll set up the providers.[Extracting architecture]Also use JWT strategy.It's a strict requirement.Updated config to JWT.[Extracting tech-context]Type a message...LanceDB Vector StoreStorage ActiveDIMENSIONS: 1024INDEX: HNSW

Everything your AI needs to remember

codexfi extracts, stores, and retrieves project knowledge automatically — so your AI agent has full context from day one.

Auto-Extraction

Automatically extracts architecture, decisions, and patterns from every conversation.

Typed Memory

10+ memory types — architecture, tech-context, progress, preferences, and more.

Semantic Search

Voyage AI embeddings find relevant context for every new task.

Compaction Survival

Memory persists through context window compaction — nothing is lost.

Zero Config

One command install. No Docker, no external database. Bring your own LLM + Voyage AI keys.

Zero Overhead

Memory fires through hooks — never a tool call, never a wasted turn. Your agent's full context budget stays free for code.

Local & Private

Memory storage is fully local. No cloud sync, no external persistence layer.

Smart Aging

Old memories are automatically consolidated and evolved, never duplicated.

How it works

Three steps. No ongoing maintenance. Your AI gets smarter with every session.

1

Install

One command. Works with OpenCode. No configuration needed.

$ curl -fsSL https://codexfi.com/install | bash
2

Code

You code. After each turn, codexfi silently extracts what matters and refreshes the memory context — no prompts, no tool calls, nothing asked of you or your agent.

3

Remember

Next session, your AI has full context. Architecture, decisions, patterns — all injected automatically.